xyt000-xjj

I use my own dataset, how can I save the model if I cannot calculate IS


What is the structure diagram of the generator like

Why is there a difference between evaluating while training and evaluating the AP obtained after saving the model after training? How to solve this problem

When I use the dataset in COCO format and pretrain the model with ImageNet, all the AP values are NaN. However, when I use a COCO-pretrained model, training proceeds normally....

How to not use pre-trained models? If I switch to networks other than the three backbone networks, do I need to use the pre-trained models trained on the COCO dataset?

Is the pre-trained model using the COCO-format dataset named something like diffdet_coco_res101.pth?